Related
Is there a way to get a list of tables that have one-on-one relationship to a given table in SQLite3?
For example, here table ab has a one-on-one relationship with both table abc and abd. Is there a query or queries to return abc and abd for the given table name ab?
-- By default foreign key is diabled in SQLite3
PRAGMA foreign_keys = ON;
CREATE TABLE a (
aid INTEGER PRIMARY KEY
);
CREATE TABLE b (
bid INTEGER PRIMARY KEY
);
CREATE TABLE ab (
aid INTEGER,
bid INTEGER,
PRIMARY KEY (aid, bid)
FOREIGN KEY (aid) REFERENCES a(aid)
FOREIGN KEY (bid) REFERENCES b(bid)
);
-- tables 'ab' and 'abc' have a one-on-one relationship
CREATE TABLE abc (
aid INTEGER,
bid INTEGER,
name TEXT NOT NULL,
PRIMARY KEY (aid, bid) FOREIGN KEY (aid, bid) REFERENCES ab(aid, bid)
);
-- tables 'ab' and 'abd' have a one-on-one relationship
CREATE TABLE abd (
aid INTEGER,
bid INTEGER,
value INTEGER CHECK( value > 0 ),
PRIMARY KEY (aid, bid) FOREIGN KEY (aid, bid) REFERENCES ab(aid, bid)
);
CREATE TABLE w (
id INTEGER PRIMARY KEY
);
The following tedious precedure may get me the list of tables I want:
Get primary keys for table ab:
SELECT l.name FROM pragma_table_info('ab') as l WHERE l.pk > 0;
get foreign keys for other tables (this case is for table abd):
SELECT * from pragma_foreign_key_list('abd');
Do parsing to get what the list of tables of one-on-one relationships.
However, there must exist a more elegant way, I hope.
For SQL Server, there are sys.foreign_keys and referenced_object_id avaible (see post). Maybe there is something similar to that in SQLite?
Edit: adding two more tables for test
-- tables 'ab' and 'abe' have a one-on-one relationship
CREATE TABLE abe (
aid INTEGER,
bid INTEGER,
value INTEGER CHECK( value < 0 ),
PRIMARY KEY (aid, bid) FOREIGN KEY (aid, bid) REFERENCES ab
);
-- tables 'ab' and 'abf' have a one-on-one relationship
CREATE TABLE abf (
aidQ INTEGER,
bidQ INTEGER,
value INTEGER,
PRIMARY KEY (aidQ, bidQ) FOREIGN KEY (aidQ, bidQ) REFERENCES ab(aid, bid)
);
Edit: verify FK for table abe
sqlite> PRAGMA foreign_keys;
1
sqlite> .schema abe
CREATE TABLE abe (
aid INTEGER,
bid INTEGER,
value INTEGER CHECK( value < 0 ),
PRIMARY KEY (aid, bid) FOREIGN KEY (aid, bid) REFERENCES ab
);
sqlite> DELETE FROM abe;
sqlite> INSERT INTO abe (aid, bid, value) VALUES (2, 1, -21);
sqlite> INSERT INTO abe (aid, bid, value) VALUES (-2, 1, -21);
Error: FOREIGN KEY constraint failed
sqlite> SELECT * FROM ab;
1|1
1|2
2|1
Alternative
Although not a single query solution the following only requires submission/execution of a series of queries and is therefore platform independent.
It revolves around using two tables:-
a working copy of sqlite_master
a working table to store the the output of SELECT pragma_foreign_key_list(?)
Both tables are created via a CREATE-SELECT, although neither has any rows copied, so the tables are empty.
A trigger is applied to the working copy of sqlite_master to insert into the table that stores the result of SELECT pragma_foreign_key_list(table_name_from_insert);
The relevant rows are copied from sqlite_master via a SELECT INSERT and thus the triggering populates the store table.
The following is the testing code :-
DROP TABLE IF EXISTS fklist;
DROP TABLE IF EXISTS master_copy;
DROP TRIGGER IF EXISTS load_fklist;
/* Working version of foreign_key_list to store ALL results of SELECT pragma_foreign_key_list invocation */
CREATE TABLE IF NOT EXISTS fklist AS SELECT '' AS child,*
FROM pragma_foreign_key_list((SELECT name FROM sqlite_master WHERE type = 'not a type' LIMIT 1));
/* Working version of sqlite master */
CREATE TABLE IF NOT EXISTS master_copy AS SELECT * FROM sqlite_master WHERE type = 'not a type';
/* Add an after insert trigger for master copy to add to fklist */
CREATE TRIGGER IF NOT EXISTS load_fklist
AFTER INSERT ON master_copy
BEGIN
INSERT INTO fklist SELECT new.name,* FROM pragma_foreign_key_list(new.name);
END
;
/* Populate master_copy from sqlite_master (relevant rows)
and thus build the fklist
*/
INSERT INTO master_copy SELECT *
FROM sqlite_master
WHERE type = 'table'
AND instr(sql,' REFERENCES ') > 0
;
SELECT * FROM fklist;
DROP TABLE IF EXISTS fklist;
DROP TABLE IF EXISTS master_copy;
DROP TRIGGER IF EXISTS load_fklist;
Using a similar test base as per the previous answer the above results in :-
Is there a way to get a list of tables that have one-on-one relationship to a given table in SQLite3?
Not with certainty as coding a Foreign Key constraint does not define a relationship (rather it supports a relationship), that is relationships can exists without a FK constraint.
A Foreign Key constraint defines:-
a) a rule that enforces referential integrity
b) optionally maintains/alters referential integrity when the referred to column is changed (ON DELETE and ON UPDATE )
As such looking at the Foreign Key List only tells you where/if a FK constraint has been coded.
Saying that the following will get the tables with the constraint and the referenced tables.
More elegant is a matter of opinion, so it's up to you :-
WITH cte_part(name,reqd,rest) AS (
SELECT name,'',substr(sql,instr(sql,' REFERENCES ') + 12)||' REFERENCES '
FROM sqlite_master
WHERE sql LIKE '% REFERENCES %(%'
UNION ALL
SELECT
name,
substr(rest,0,instr(rest,' REFERENCES ')),
substr(rest,instr(rest,' REFERENCES ') + 12)
FROM cte_part
WHERE length(rest) > 12
)
SELECT DISTINCT
CASE
WHEN length(reqd) < 1 THEN name
ELSE
CASE substr(reqd,1,1)
WHEN '''' THEN substr(replace(reqd,substr(reqd,1,1),''),1,instr(reqd,'(')-3)
WHEN '[' THEN substr(replace(replace(reqd,'[',''),']',''),1,instr(reqd,'(')-3)
WHEN '`' THEN substr(replace(reqd,substr(reqd,1,1),''),1,instr(reqd,'(')-3)
ELSE substr(reqd,1,instr(reqd,'(')-1)
END
END AS tablename
FROM cte_part
;
As an example of it's use/results :-
screenshot from Navicat
Here's an adaptation of the above that includes, where appropriate, the child table that references the parent :-
WITH cte_part(name,reqd,rest) AS (
SELECT name,'',substr(sql,instr(sql,' REFERENCES ') + 12)||' REFERENCES '
FROM sqlite_master
WHERE sql LIKE '% REFERENCES %(%'
UNION ALL
SELECT
name,
substr(rest,0,instr(rest,' REFERENCES ')),
substr(rest,instr(rest,' REFERENCES ') + 12)
FROM cte_part
WHERE length(rest) > 12
)
SELECT DISTINCT
CASE
WHEN length(reqd) < 1 THEN name
ELSE
CASE substr(reqd,1,1)
WHEN '''' THEN substr(replace(reqd,substr(reqd,1,1),''),1,instr(reqd,'(')-3)
WHEN '[' THEN substr(replace(replace(reqd,'[',''),']',''),1,instr(reqd,'(')-3)
WHEN '`' THEN substr(replace(reqd,substr(reqd,1,1),''),1,instr(reqd,'(')-3)
ELSE substr(reqd,1,instr(reqd,'(')-1)
END
END AS tablename,
CASE WHEN length(reqd) < 1 THEN '' ELSE name END AS referrer
FROM cte_part
;
Example of the Result :-
the artists table is referenced by albums as the SQL used to create the albums table is CREATE TABLE 'albums'([AlbumId] INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL ,[Title] TEXT NOT NULL ,[ArtistId] INTEGER NOT NULL , FOREIGN KEY ([ArtistId]) REFERENCES 'artists'([ArtistId]))
i.e. FOREIGN KEY ([ArtistId]) REFERENCES 'artists'([ArtistId]))
the employees table is self-referencing as per CREATE TABLE 'employees'(.... REFERENCES 'employees'([EmployeeId]))
Additional re comment:-
(I am still trying to understand your code...)
The code is based upon selecting rows from sqlite_master where the row is for a table (type = 'table'), as opposed to an index, trigger or view and where the sql column contains the word REFERENCES with a space before and after and there is a following left parenthesis.
The last condition used to weed out the likes of CREATE TABLE oops (`REFERENCES` TEXT, `x REFERENCES Y`);
For each selected row 3 columns are output:-
name which is the name of the table as extracted from the name column of sqlite_master,
reqd is initially an empty string (i.e. initial)
rest the rest of sql that follows the referred to table name with suffixed with REFERENCES.
The UNION ALL adds rows that are built upon what is newly added to the CTE, i.e. the three columns are extracted as per :-
name is the name
reqd is the sql from the rest column up until the first REFERENCES term (i.e. the table and referenced column(s))
rest is the sql from after the REFERENCES term
As with any recursion the end needs to be detected, this is when the entire sql statement has been reduced to being less than 12 (i.e the length of " REFERENCES ", the term used for splitting the sql statement).
This is what is termed as a RECURSIVE CTE
Finally the resultant CTE is then queried. If the reqd field is empty then the tablename column is the name column otherwise (i.e. the reqd column contains data(part of the sql)) the table name is extracted (part up to left parenthesis if not enclosed (`,' or [ with ])) or extracted from between the enclosure.
The following is what the final query results in if all the CTE columns are included (some data has been truncated):-
As can clearly be seen the extracted sql progressively reduces
The answer is intended as in-principle and has not been extensively tested to consider all scenarios, it may well need tailoring.
CREATE TABLE Client_master (
Client_no varchar(6) PRIMARY KEY,
Name varchar(15) NOT NULL,
City varchar(15),
Pincode number(8),
State varchar(15),
Bal_due Number(10,2),
CHECK(Client_no LIKE 'C%'));
INSERT INTO
Client_master(Client_no,Name,City,Pincode,State,Bal_due)
VALUES('C00001','Ivan Bayross','Bombay','400054','Maharashtra',15000);
1 row created.
CREATE TABLE Sales_order(
Order_no varchar(6) PRIMARY KEY REFERENCES Client_master (Client_no),
Order_date date,
Client_no varchar(6),
Dely_type char(1) DEFAULT 'f',
Billed_yn char(1),
Salesman_no varchar(6),
Dely_date date,
Order_status varchar(10),
CHECK(Order_no LIKE 'O%'),
CHECK(Order_status IN ('inprocess','backorder','cancelled')),
CHECK(Dely_date>Order_date));
INSERT INTO Sales_order(Order_no, Order_date, Client_no, Dely_type, Billed_yn, Salesman_no, Dely_date, Order_status)
VALUES('C19001','12-jan-96','C00001','f','n','S0001','20-jan-96','inprocess');
INSERT INTO
*
ERROR at line 1:
ORA-02291: integrity constraint (SYSTEM.SYS_C007155) violated - parent key not
found
Please help me insert data into Child table.
What's this SYSTEM.SYS_C007155 error? Why this error message "parent key is not found"?
The parent key not found is because you have a foreign key in your sales_order table which is pointing to the client_master, and there is no matching key value - exactly what it says. But it's because you've done something odd:
Sales_order(Order_no varchar(6) PRIMARY KEY REFERENCES Client_master (Client_no)
You've made the order_no the primary key for this table, but also made it a foreign key to the client_no on the other table. Your insert uses order_no of 'C19001', which doesn't match the client_no you previously inserted into the parent table.
You almost certainly wanted sales_order.client_no to be a foreign key to client_master.client_no, so you would have the references... against that column:
CREATE TABLE Sales_order(Order_no varchar(6) PRIMARY KEY,
Order_date date,
Client_no varchar(6) REFERENCES Client_master (Client_no),
...
You have a further problem as the check constraint you have against the order_no is CHECK(Order_no LIKE 'O%') but as mentioned the value you're putting in is 'C19001', which doesn't match the pattern. Presumably you meant the insert to be for 'O19001'; if not then the constraint is defined incorrectly. I actually hit that check constraint before the primary key constrain in 11gR2, so you may just have changed that while posting the question.
You can look in the user_constraints and user_cons_columns views to see what a constraint is doing, but you'll find it easier if you name your constraints instead of letting Oracle give them default names like SYS_C007155:
Specify a name for the constraint. If you omit this identifier, then Oracle Database generates a name with the form SYS_Cn. Oracle stores the name and the definition of the integrity constraint in the USER_, ALL_, and DBA_CONSTRAINTS data dictionary views (in the CONSTRAINT_NAME and SEARCH_CONDITION columns, respectively).
For example:
CREATE TABLE Client_master (
Client_no varchar(6),
Name varchar(15) NOT NULL,
City varchar(15),
Pincode number(8),
State varchar(15),
Bal_due Number(10,2),
CONSTRAINT Client_master_pk PRIMARY KEY (Client_no),
CONSTRAINT Client_master_chk_no CHECK(Client_no LIKE 'C%'));
INSERT INTO Client_master(Client_no,Name,City,Pincode,State,Bal_due)
VALUES('C00001','Ivan Bayross','Bombay','400054','Maharashtra',15000);
1 row inserted.
CREATE TABLE Sales_order (
Order_no varchar(6),
Order_date date,
Client_no varchar(6),
Dely_type char(1) DEFAULT 'f',
Billed_yn char(1),
Salesman_no varchar(6),
Dely_date date,
Order_status varchar(10),
CONSTRAINT Sales_order_pk PRIMARY KEY (Order_no),
CONSTRAINT Sales_order_fk_client FOREIGN KEY (Client_no)
REFERENCES Client_master (Client_no),
CONSTRAINT Sales_order_chk_no CHECK (Order_no LIKE 'O%'),
CONSTRAINT Sales_order_shk_status CHECK
(Order_status IN ('inprocess','backorder','cancelled')),
CONSTRAINT Sales_order_chk_dates CHECK(Dely_date>Order_date));
INSERT INTO Sales_order(Order_no, Order_date, Client_no, Dely_type, Billed_yn, Salesman_no, Dely_date, Order_status)
VALUES('O19001','12-jan-96','C00001','f','n','S0001','20-jan-96','inprocess');
1 row inserted.
If you do have a violation, e.g. with your original order_no value, you'd see a more useful message:
INSERT INTO Sales_order(Order_no, Order_date, Client_no, Dely_type, Billed_yn, Salesman_no, Dely_date, Order_status)
VALUES('C19001','12-jan-96','C00001','f','n','S0001','20-jan-96','inprocess');
ORA-02290: check constraint (YOUR_SCHEMA.SALES_ORDER_CHK_NO) violated
so you can have a better idea which constraint is violated, and what that means, without having to loo in the data dictionary. SALES_ORDER_CHK_NO is easier to interpret than SYS_C007155 Use names that make sense for you of course.
You can name inline constraints too:
CREATE TABLE Client_master (
Client_no varchar(6) CONSTRAINT Client_master_pk PRIMARY KEY,
...
CREATE TABLE Sales_order (
Order_no varchar(6) CONSTRAINT Sales_order_pk PRIMARY KEY,
Order_date date,
Client_no varchar(6) CONSTRAINT Sales_order_fk_client REFERENCES Client_master (Client_no),
...
but it might be clearer and easier to maintain with them all grouped together at the end.
Original table1 and Table2. Both tables has data.
CREATE TABLE [dbo].[Table1]
(
[Id] int NOT NULL PRIMARY KEY
)
CREATE TABLE [dbo].[Table2]
(
[Id] INT NOT NULL PRIMARY KEY,
[Table1Id] Int NULL,
Constraint [FK_Table1_Table2] foreign key ([Table1Id]) references [Table1] (Id)
)
I'd like to change the Table1.Id to UNIQUEIDENTIFIER.
Obviously just jump in and change the type from int to UNIQUEIDENTIFIER for Table1.Id and 'Table2.Table1Id'. Then Publish. Here is the code:
CREATE TABLE [dbo].[tmp_ms_xx_Table1] (
[Id] UNIQUEIDENTIFIER NOT NULL,
PRIMARY KEY CLUSTERED ([Id] ASC)
);
IF EXISTS (SELECT TOP 1 1
FROM [dbo].[Table1])
BEGIN
INSERT INTO [dbo].[tmp_ms_xx_Table1] ([Id])
SELECT [Id]
FROM [dbo].[Table1]
ORDER BY [Id] ASC;
END
This code will fail because original Table1.Id is Int while temp table Id is UNIQUEIDENTIFIER.
Then, I try to with Pre-Scripts. Ideally all the changes will be done manually.
--drop fk constraint
alter table [Table2] drop constraint [FK_Table1_Table2];
--rename table1.id
exec sp_rename 'Table1.Id', 'Id2', 'COLUMN';
alter table [Table1] add Id uniqueidentifier not null
default newid();
--rename table2.table1id
exec sp_rename 'Table2.Table1Id', 'Table1Id2', 'COLUMN';
alter table [Table2] add Table1Id uniqueidentifier null;
update t2 set t2.Table1ID = t1.Id
from Table2 t2 left join Table1 t1 on t2.Table1Id2 = t1.Id2;
alter table [Table2] add constraint [FK_Table1_Table2] foreign key (Table1Id) references Table1 (Id);
However it FAIL again as SSDT is trying to compare its data structure again the target database.
Any idea please?
You're right. The problem is that you can't include schema changes in the pre-deployment script because SSDT's deployment script is generated prior to your schema changes. It is therefore only useful for data-only changes.
The solution is to do this outside of the SSDT process altogether. Yes, it's a pre-pre-deployment script! Essentially you have to apply your change by yourself before you even get to the SSDT bit.
(There's probably a way to do this via a custom deployment contributor. After all, everything is possible in code...)
Can I convince you to take a look at a migration-based solution as it appears that you have sufficient need for an element of fine-grained script "customisation". DBUp is a popular open source solution. ReadyRoll is a more-integrated commercial solution that shares a lot with SSDT.
the problem is the old data. it will be ok without the data in table in step 2.
1.pre-script: copy/process old data to temp tables, delete them from original tables
create table #table1 (
id int null,
id2 uniqueidentifier null
);
insert into #table1 (id,id2)
select id,newid() from Table1;
create table #table2 (
id int null,
table1id int null,
table1id2 uniqueidentifier null
);
insert into #table2 (id, table1id)
select id,table1id from Table2;
update t2 set t2.table1id2=t1.id2
from #table2 t2 left join #table1 t1 on t2.table1id = t1.id;
delete from table2;
delete from table1;
dacpac will auto generate the changes for schema. it will be ok because no data is existing any more.
post-script: insert data back from temp tables in pre-script:
insert into table1 (id)
select id2 from #table1;
insert into table2 (id,table1id)
select id, table1id2 from #table2;
The Alembic migration for a SQLite database:
def upgrade():
with op.batch_alter_table('my_table') as batch_op:
batch_op.add_column(sa.Column('parent_id', sa.String(24)))
batch_op.create_foreign_key('parent_constraint', 'my_table', ['parent_id'], ['id'])
which is supposed to create a foreign key parent_id referencing id of the same table my_table, creates a reference to a table called _alembic_batch_temp:
CREATE TABLE "my_table" (
id VARCHAR(24) NOT NULL,
parent_id VARCHAR(24),
PRIMARY KEY (id),
CONSTRAINT parent_constraint FOREIGN KEY(parent_id) REFERENCES _alembic_batch_temp (id)
)
How to create self-referencing constraints when altering a table?
After some research I found that the problem here is the way Alembic does the batch migration. In short, at the current version (0.7.6) of Alembic it's not possible to create relation with self by migration.
As described in the Alembic documentation, to do the migration, new table is created with a temporary name and changes from the alter table
code. In this case:
CREATE TABLE _alembic_batch_temp (
id VARCHAR(24) NOT NULL,
parent_id VARCHAR(24),
PRIMARY KEY (id),
CONSTRAINT parent_constraint FOREIGN KEY(parent_id) REFERENCES _alembic_batch_temp (id)
)
The table is filled with the data from the old table:
INSERT INTO _alembic_batch_temp (id) SELECT id FROM my_table;
Then the old table is removed:
DROP TABLE my_table;
Finally the newly created table is renamed to it's proper name:
ALTER TABLE _alembic_batch_temp RENAME TO my_table;
The problem with this way of doing things is already visible in the first code snippet. The newly created foreign key is referencing the temporary table and once it's created it can't be changed due to restrictions in SQLite. So after the renaming of the table you end up with the table you provided:
CREATE TABLE "my_table" ( # new name
id VARCHAR(24) NOT NULL,
parent_id VARCHAR(24),
PRIMARY KEY (id),
CONSTRAINT parent_constraint FOREIGN KEY(parent_id) REFERENCES _alembic_batch_temp (id) # old reference
)
To Avoid this situation you can create the batch migration manually:
Rename the old table to some temporary name:
ALTER TABLE my_table RENAME TO migration_temp_table;
Create new table with proper name and proper reference:
CREATE TABLE my_table (
id VARCHAR(24) NOT NULL,
parent_id VARCHAR(24),
PRIMARY KEY (id),
CONSTRAINT parent_constraint FOREIGN KEY(parent_id) REFERENCES my_table (id)
)
Copy the data:
INSERT INTO my_table (id) SELECT id FROM migration_temp_table;
Remove the old table:
DROP TABLE migration_temp_table;
I have the following 5 tables defined with a few records inserted into the 1st 4. This is using sqlite 3.7.1.7 with foreign key constaint enabled.
create table if not exists subject (id varchar(50) primary key,desc varchar(100));
insert into subject (id,desc) values ("subject1","test subject");
create table if not exists subjectlevel (id_subject_id varchar(50) references subject(id) on delete cascade, id integer not null, desc varchar(100) not null, questmcmaxselections integer not null, primary key (id_subject_id,id));
insert into subjectlevel (id_subject_id,id,desc,questmcmaxselections) values ("subject1",1,"test subject1 level 1",4);
insert into subjectlevel (id_subject_id,id,desc,questmcmaxselections) values ("subject1",2,"test subject1 level 2",4);
create table if not exists questmc (id integer primary key, text varchar(300) not null, includeallanswers int not null, subject_id varchar(50), subjectlevel_id integer, foreign key (subject_id, subjectlevel_id) references subjectlevel (id_subject_id,id) on delete cascade);
insert into questmc (text,includeallanswers,subject_id,subjectlevel_id) values ("this is a _ question", 1, "subject1",1);
create table if not exists questmcselection (id integer primary key, text varchar(100) not null, subject_id varchar(50), subjectlevel_id integer, foreign key (subject_id, subjectlevel_id) references subjectlevel (id_subject_id,id) on delete cascade);
insert into questmcselection (text,subject_id,subjectlevel_id) values ("this is a solution","subject1",1);
create table if not exists questmc_questmcselection(id integer primary key, answer integer not null, questmc_id integer, questmcselection_id integer, subject_id varchar(50), subjectlevel_id integer, foreign key (questmc_id) references questmc(id) on delete cascade, foreign key (questmcselection_id) references questmcselection (id) on delete cascade, foreign key (subject_id,subjectlevel_id) references questmc (subject_id,subjectlevel_id) on delete cascade, foreign key (subject_id,subjectlevel_id) references questmcselection (subject_id,subjectlevel_id));
if i attempt to delete the second record in the subjectlevel table, i get a foreign key mismatch error as long as table questmc_questmcselection is defined.
sqlite> delete from subjectlevel where id=2;
Error: foreign key mismatch - "questmc_questmcselection" referencing "questmcselection"
questmc, questmcselection, and questmc_questmcselection have no related existing records that should prevent this deletion. Any idea why this error occurs?
This error has nothing to do with this particular subjectlevel record.
Your problem is that your tables lack the required indexes.
This was not reported earlier because that DELETE statement was the first command that required SQLite to check the consistency of the database schema.
Based on CL's answer -
sqlite> create table parent(a);
sqlite> create table child(a, FOREIGN KEY (a) REFERENCES parent(a));
sqlite> pragma foreign_keys = ON;
sqlite> insert into parent values(3);
sqlite> insert into child values (3);
Error: foreign key mismatch - "child" referencing "parent"
sqlite> create unique index p_a on parent(a);
sqlite> insert into child values (3);
sqlite> _
From the documentation:
Usually, the parent key of a foreign key constraint is the primary key
of the parent table. If [not], then the parent key columns must be
collectively subject to a UNIQUE constraint or have a UNIQUE index [which uses]
the collation sequences ... in the CREATE TABLE
statement for the parent table.
i.e. the alternative is:
sqlite> create table parent(a, b, UNIQUE (a, b));
sqlite> create table child (x, y, FOREIGN KEY (x, y) REFERENCES parent(a, b));
(this also highlights multi-column foreign keys; they work with indexes too...)